Fig. 3: ML Model Feature Importances.
From: Learning grain boundary segregation behavior through fingerprinting complex atomic environments
Top six feature SHAP impacts for each ML model a STGB, b Amorphous, and c 6nm sorted by mean absolute SHAP value. The SHAP value of each point indicates the effect of the indicated feature on the final model prediction for that atom48. When correlated with changes in the feature value across the model, this indicates the effect of changes in the feature value on the output value (Eseg). X-axis indicates the SHAP value for each point, Y-axis indicates which SFD feature is being considered, and coloring indicates the relevant feature value. Y-axis deviation along each feature row indicates areas of higher and lower density for the corresponding feature and SHAP value.
